World champion in automatic image and video search
21 October 2009
Image and video search system wins 3 prestigious international competitions
How do you find an image or video on the Internet or YouTube without a text description? Along with his colleagues at the University of Amsterdam, Dutch researcher Theo Gevers has developed software that can find objects and concepts automatically in images and video clips. The research team has achieved a unique feat: they’ve scooped first prizes in three international competitions for their work. Theo Gevers received a “Vici” grant from NWO’s Innovational Research Incentives Scheme in 2007 for his groundbreaking work.
The team from the University of Amsterdam took part in three major international competitions, testing and comparing computer search engines. These competitions looked into which search methods could best detect concepts in images and video clips without text descriptions. Concepts include objects such as persons, animals, vehicles and events such as explosions, demonstrations, fire, violence etc. The team scooped first prize in three different international competitions (with over 60 teams from both industry and academia), including ImageCLEF and PASCAL VOC. The world-leading image and video search system uses advanced automatic image understanding, rapid indexing and intelligent classification techniques. Supercomputers are used to address the enormous volumes of data.
Computerised searches
How do you look for an action scene or a pretty cat? At the moment, we use text for this. For instance, you can type in the word “cat” and a search engine will start to look for images linked to the word “cat”. But if the image or film clip is not described by text, then searching against text becomes impossible. The team from University of Amsterdam has changed this by using an automatic computerised search for objects within the image itself, rather than for the text.
Applications
The growing number of video files means there is a growing need for search engines that facilitate searches for pictures and video fragments. Computerised object identification makes it easier, for example, to search through large databases like YouTube, Flickr and Facebook for features not included in the text.
..............................
For further information please contact:
- Theo Gevers (University of Amsterdam)
- tel: +31 (0)20 525 7516
- e-mail: Th.Gevers@uva.nl
